Color-FaaS

Color FaaS Design Doc

Designed By Hanzhong Liu, Xi Shi

Video: https://www.youtube.com/watch?v=QDuIaaMpTBo

High Level Design

Requirements

I plan to design a FaaS platform called Color FaaS. Color FaaS will has the ability to process many kinds of task like AWS Lambda. The platform will include user registration, function creation and management, billing and payment.

User case 1: Development

Items
Actor	Developer
Goal	A User want to deploy a application in cloud environment
Path	1. User must develop a func in his/her computer by a SDK and upload this function from the webpage 2. System saves the function file, and the function file is successfully created 3. User test this function from web browser. 4. System calls the function and returns the result

User case 2: Pay the bill

Items
Actor	Developer
Goal	Pay the bill
Path	1. User login into the platftom 2. System verifies that user information is correct 3. User enter the account info page 4. System return the web page to user 5. User click “bill” button 6. System queries all unpaid records and generates a bill. Platform will show the bill in web page 7. User click “Pay” button 8. System redirects to the payment page 9. User enter all information in the payment form and submit 10. System deducts the fee and completes the payment

User case 3: Manage function

Items
Actor	Developer
Goal	Trace the running status of a Func
Path	1. User login into the platform 2. System verifies that user information is correct 3. User enter Task manage page 4. System return the manage page 5. User use search to find the target Func 6. System search from the database and return it to user 7. User look at the running status of the function

Architectural design

The system is divided into three layers: the website, the gateway, and the cluster. The website is primarily responsible for user interaction and needs to connect to the database to store user information. The gateway layer is a simplified implementation that can be replaced with gateways like AWS ELB (Elastic Load Balancer). The cluster is the core of the entire Color FaaS platform. Here, I have designed a decentralized resource scheduler and use Docker for handling user tasks.

In Website layer, we use Java Serverlet to develop a Web application.We also use Mysql to store user’s information. Website will provide user with functions include: add/delete a Func, pay for the bill, trace Func status.

For GateWay, all request will be send to gateway, and gateway will write infomation about the task into DB. In the end, gateway will sent task to Core cluster.

Core cluster may contains many servers, so it can process many tasks in the same time. A worker program will run in In these servers. When a task is sent to a worker, it will stand up a Docker container and run this task in that container. Because, image of a container is very big, so worker will use LRU to cache some containers.

Website Detail Design

UI Design / Prototype

- Login page

- Register page

- Func manage page

From this page, user can Add, delete, or modify a function.

- Task manage page

This page shows all the user’s functions running.


- Account info page

This page shows the bill and account balance, and the user can pay the bill or top up the money.

Database Design

- User Info Table:

Item	Type	Explanation
User_ID	Bigint
User_Email	varchar(50)
User_Name	Varchar(20)
User_Password	varchar(20)
User_Balance	Bigint

- Bill Table

Item	Type	Explanation
Bill_ID	Bigint
User_ID	Bigint	Index
Bill_Date	Datetime
Bill_Balance	Bigint
Bill_Detail	Text	A JSON inclueds all unpaid function running records’ ID {1,2,3}
Status	int	0 unsay, 1 paid

- Function Table

Item	Type	Explanation
Func_ID	Bigint
User_ID	Bigint
Func_Args	JSON	{“num1”:int,”num2”,int}
Func_Path	varchar(200)	path of func’s code in OSS like S3
Func_Explanation	varchar(500)
Func_Output	Int	0=sync / 1=async
Func_Resource	JSON	{“CPU”:4,”MEM”:1024m,”DISK”:1024m}

- Func Running Record

tem	Type	Explanation
Running_ID	Bigint
Func_ID	Bigint
Running_Time	Bigint	Example: 100ms
Status	Int	0: fail, 1 success, 2 pending
Input	Text	A input json
Running_Date	Datetime
Payment_Status	Int	0 unsay, 1 paid

- Billing Unit

Item	Type	Explanation
ID	Bitint
Resource_Type	Int	CPU / MEM / DISK
Price	float	Example: 0.01$/ms

- Account Info History

Item	Type	Explanation
ID	Bitint
Date	Datetime	CPU / MEM / DISK
Type	Int
Account_num	varchar(20)
Bill_ID	Int
Money	Bigint
User_ID	Bigint

- ER

- SQL Code

create table Account_Info_History
(
    ID          bigint auto_increment
        primary key,
    Date        varchar(30) null,
    Type        int         null comment '1:money in, pay for a bill',
    Account_Num varchar(20) null comment 'for type 1',
    Bill_ID     bigint      null comment 'for type 2',
    Money       bigint      null comment 'for type 1, it is a positive num.',
    User_ID     bigint      null
);

create table Bill
(
    Bill_ID      bigint auto_increment
        primary key,
    User_ID      bigint   null,
    Bill_Date    datetime null,
    Bill_Balance double   null,
    Bill_Detail  text     null,
    Status       int      null comment '"0 unpay, 1paid"'
);

create index User_ID
    on Bill (User_ID);

create table Billing_Units
(
    ID            bigint auto_increment
        primary key,
    Resource_Type varchar(10) null,
    Price         double      null comment 'Example: 0.01$/ms'
);

create table Func_Running_Record
(
    Running_ID     bigint auto_increment
        primary key,
    Func_ID        bigint   null,
    Running_Time   int      null comment '100ms',
    Status         int      null comment '0: fail, 1 success, 2 pending',
    Input          text     null,
    Running_Date   datetime null,
    Payment_Status int      null comment '"0 unpay, 1 paid"'
);

create table Functions
(
    Func_ID            bigint auto_increment
        primary key,
    User_ID            bigint       null,
    Func_Args          text         null,
    Func_Path          varchar(300) null,
    Func_Explanation   varchar(500) null,
    Func_Output        int          null comment '0=sync / 1=async',
    Func_Resource_CPU  int          null,
    Func_Resource_Mem  int          null,
    Func_Resource_Disk int          null,
    Func_Name          varchar(50)  null
);

create index User_ID
    on Functions (User_ID);

create table User_Info
(
    User_ID       bigint auto_increment
        primary key,
    User_Email    varchar(50) null,
    User_Password varchar(50) null,
    User_Balance  bigint      null
);

Core Cluster Detail Design

**In development! **

In core cluster, there are many servers. All servers will be managed by zookeeper. All nodes can perform the scheduling function. Zookeeper will elect a leader for scheduling.

The worker nodes will regularly report resource usage to the master node, and the master node will use the progress of DRF algorithm based on resource usage information

The following figure shows the module division of the whole scheduling system(For master):